Chunk OrthogonalProcrustesAlignment top_k search to cut peak memory

[AmitSubhash]

Closes #307.

Problem

OrthogonalProcrustesAlignment.fit materializes the full (n_label, n_ref) distance matrix and its row-wise argsort before selecting the top_k closest references. subsample is only applied afterwards, so it cannot reduce peak memory: peak stays O(n_label * n_ref) no matter how small subsample is. That is the blow-up @onford reported.

Fix

Run the top_k search in blocks of label rows. Each block is scored against all references, so the per-row selection is unchanged while peak memory drops to O(block * n_ref).

Because the chunking is over rows only, never over the feature/contraction dimension, each row's distance vector is computed exactly as before. The result is therefore bit-identical to the previous implementation, ties included.

The block size is internal (derived from a module-level constant), so there is no public API change. I'd floated a chunk_size constructor parameter in #307, happy to add it if you'd prefer, but the fix doesn't need user tuning so I kept the surface minimal.

Measured effect

15000 x 15000 alignment (top_k=5, subsample=1000), peak RSS:

path	peak RSS
previous (full matrix)	5372 MB
this PR	621 MB

roughly 8.6x lower.

Tests

Added to tests/test_data_helper.py:

• test_orthogonal_alignment_chunking_matches_full — parametrized over shapes, top_k (including top_k == n_ref) and 1D/2D labels. Compares against an independently computed full-matrix reference, then asserts the multi-block and single-block runs are bit-identical (assert_array_equal) for both _transform and transform(data).
• test_orthogonal_alignment_chunking_bit_identical_with_ties — duplicated integer labels give tied distances; asserts bit identity.
• test_orthogonal_alignment_chunking_bounds_block_size — asserts the search really runs in bounded blocks (spies on _distance). I checked that reverting to the old full-matrix path makes only this test fail, so it guards the memory contract rather than just re-testing equivalence.

pytest tests/test_data_helper.py -m "not requires_dataset" (46 passed) and the helper.py doctests pass; yapf and isort are clean.

[cla-bot]

Thank you for your contribution. We require contributors to sign our Contributor License Agreement (CLA). We do not have a signed CLA on file for you. In order for us to review and merge your code, please sign our CLA here. After you signed, you can comment on this PR with @cla-bot check to trigger another check.

[AmitSubhash]

@cla-bot check

[cla-bot]

Thanks for tagging me. I looked for a signed form under your signature again, and updated the status on this PR. If the check was successful, no further action is needed. If the check was unsuccessful, please see the instructions in my first comment.